date: Wednesday 9/12/2020
- Stefano Cozzini : Network Hardware and Performance Model to evaluate MPI applications
- chapter 9 of reference 4: final section: code here provided is taken from the book's website
- chapter 10 of reference 4: final section contains interesting stuff.
It is a "simple" fortran implementation of the Jacobi solver in 3D
To compile, load the MPI module and then:
mpif77 -ffixed-line-length-none Jacobi_MPI_vectormode.F -o jacoby3D.x
To run the code
mpirun --mca btl ^openib -np 24 ./jacoby3D.x
get the benchmark:
git pull https://github.com/intel/mpi-benchmarks
Enter in the dir:
cd mpi-benchmarks/src_c
Compile the code, once you load the module:
make
Run the executable across two nodes:
mpirun -np 2 --map-by node --report-bindings ./IMB-MPI1 PingPong
Run the executable across two socket:
mpirun -np 2 --report-bindings --map-by socket ./IMB-MPI1 PingPong
Run the executable within the same socket:
mpirun -np 2 --report-bindings --map-by socket ./IMB-MPI1 PingPong
Run the executable using a different protocol: tcp and not infiniband
mpirun -np 2 --mca pml ob1 --mca btl tcp,self --mca btl_tcp_if_include br0 ./IMB-MPI1 PingPong